Question on LRGB Workflow re: timing of StarAlignment and DBE

[javajunkie2121]

Hi:

would appreciate some advice.

I'm moving from OSC to mono LRGB imaging, and trying to learn the new workflow. I've looked at some tutorial examples..they differ a bit in details regarding the timing of registration, or number of times you register stacks, as well as when to execute DBE

1. Let's say I've got binned 1x1 luminance, but binned 2x2 R, G and B frames, so RGB in need of upsampling. 

It's been suggested in the IP4AP series that once you calibrate L-R-G-B frames to their respective flats with darks/bias etc..that you register all individual L-R-G-B together in one big batch, then separately integrate the different channels. Is this superior to separately registering and integrating the L, then R, G, and then B into their respective stacks? 

If you register all the frames together in one big batch, and then separately integrate, I presume the R-G-B were already upsampled if you used the luminance as the reference.  If the next step is to combine the R, G abd B channels, do you need to reuse Star align if you've already registered the images together?  especially if you've applied the same dynamic crop instance to L-R-G-B stacks to maintain registration?

Do you need downstream to reuse staralignment again to combine the L with the RGB composite?  or would the LRGBCombination tool maintain the registration?

2. should you do DBE on each individual R, G and B master first, and then combine the channels into the RGB composite, or should you combine the channels into a composite RGB first, and do DBE on the composite? 

jeff

[javajunkie2121]

anyone have any advice?

jeff

[RickS]

1. It's best to pick a single registration master and register all your subs against that one.  It doesn't really matter if you register them all at once or in batches.  As you mention, you'd use a Lum sub for the master if your RGB are binned.  You shouldn't need to register again once you've done this so long as you do the same cropping, etc. on all your masters.

2. I generally DBE the individual colour masters separately but some folks do it on the combined RGB.  I don't think either method is intrinsically better.

Cheers,
Rick.

[chris.bailey]

Jeff

Registering all the LRG&B subs against a single frame just saves having to re-register the LRG&B stacks against each other prior to doing the LRGB combine. If you dither, it makes some sense to make the first frame the reference as this is the centre of the dither pattern. You then spread the misalignment around all four borders.

I find doing DBE on the RGB combine better as it lets you see if you have got rid of subtle colour gradients in the result though for mosaics I tend to mosaic LRG&B separately so doing DBE on the separate stacks can help get rid of differing gradients on the overlaps.

Chris

[RickS]

Registering all the LRG&B subs against a single frame just saves having to re-register the LRG&B stacks against each other prior to doing the LRGB combine. If you dither, it makes some sense to make the first frame the reference as this is the centre of the dither pattern. You then spread the misalignment around all four borders.

Some capture programs use a fixed pattern for dither starting at a centre point but others use a random dither.  If it's random the you're better off picking the best quality sub to register against.  Even if it's not random I suspect this is still the best strategy.

Cheers,
Rick.

[javajunkie2121]

thanks for the advice chris and rick

jeff

[Warhen]

Jeff, As I understand it, each time you align, you may be introducing artifacts as the frames are effectively 'warped' to match star centroids precisely. As Rick intimates, it is better to align once, and if binning color (I don't recommend this), to a good quality L reference frame so the binned R-G-Bs are upsampled.

Regarding DBE/ABE, I find myself following my partner RBA's advice http://www.ip4ap.com/tutorials.htm and applying BackgroundModelization to individual frames, and again to the combined RBG.

[JoLo]

Can't answer the binning question, i do all 1x1 images.  I have also taken to not shooting luminance at all, and using ChannelExtract in the CIE Lab mode for a good lum file.  I find this method produces a lower noise, higher SNR lum file to process than shooting and stacking individual lum frames, plus i can devote more time to RGB (and Ha, if shooting).

Occasionally, i run into registration problems when i try to combine all images to a single reference.  I register and integrate my R, B, G (and Ha) stacks individually, then re-register the masters to each other and the synthetic lum file.  A single Dynamic Crop on all masters and i am good to go.

For DBE, i use Warren's technique from his (great) tutorials and apply DBE to the individual frames.  I will do a test background model on the RGB and apply if it appears to help....sometimes it does not improve what has already been subtracted and I skip it.

Joe

[Zocky]

I often bin color to speed up the process.
First of all, if you bin color, you need high quality Luminance. In processing, calibration is than to each channel with their respective flats with darks/bias etc.. Than you register all of them together, with L as a reference, and in that process RGB will automatically upsample. Next is combining RGB, histogram stretch, etc. Do the histogram stretch and the rest of processing on Luminance, and finally do the LRGBCombination to merge RGB with L.

[RickS]

I often bin color to speed up the process.

Hi Zocky,

If you're using the SBIG camera in your signature then you're probably not getting much benefit from hardware binning.  The KAF-8300 does give a reduction in read noise when binning but it's not as big as one might hope (ideally the read noise for a binned x2 "super pixel" would be the same as a single unbinned pixel but it's greater than this.)  You can probably do as well with software binning or just some noise reduction at small scales.  Capturing the colour unbinned also gives you the option of extracting additional luminance from it.

Just offering some food for thought.  By all means, keep binning if it works for you.

Cheers,
Rick.

[Zocky]

Rick, I appreciate any advice. Due to constantly bad weather I never had the opportunity to perform a real test. Visually it looks like there is much more data on bin image. My theory is, that L carries sharpness and details, and the bin RGB carries only color info, so due to that they can be in lower resolution. You can find so much pros and cons on the web, and I don`t know what to think. Am I wrong?

[RickS]

Rick, I appreciate any advice. Due to constantly bad weather I never had the opportunity to perform a real test. Visually it looks like there is much more data on bin image. My theory is, that L carries sharpness and details, and the bin RGB carries only color info, so due to that they can be in lower resolution. You can find so much pros and cons on the web, and I don`t know what to think. Am I wrong?

It is a topic that has been hotly debated in many places including this forum   If you have time you might want to dig through threads like this one: http://pixinsight.com/forum/index.php?topic=6042.0

There are two benefits to binning.  One is from dividing the same number of detected photons into fewer pixels.  This increase in the number of photons per pixel results in reduced shot noise and hence higher SNR.  You don't have to do this in the camera.  You can bin in PixInsight afterwards if you wish and get the same benefit.  The second benefit is a reduction in read noise that you only get from binning in the camera.  My point is that this benefit is not realized terribly well in practice with the KAF-8300 sensor.  Also, if you're taking sky limited exposures the read noise isn't significant anyway.

So, if I were you I wouldn't bother to bin the colour.  I'd capture it unbinned and then I'd extract additional luminance information from the RGB data (see the thread I referenced for Juan's advice on how to do this).  If my colour data was noisy then I'd hit it with a fairly heavy noise reduction and rely on the enhanced luminance to supply the fine detail (as per your theory!)  You could software bin the colour data if you prefer but I think that doing noise reduction with an emphasis on small scales will produce a better result.

Hope that helps.  This is generally how I capture and process LRGB images since I also have a camera which doesn't give much read noise advantage when binning (KAF-16803 sensor.)

Cheers,
Rick.

[Zocky]

I will go through that thread and investigate deeper on this subject.
Thnx a lot Rick.

[jkmorse]

Zocky,

I was in on those threads and was strictly a binned RGB, unbinned Lum guy until my discussion with Juan.  As far as I can tell, the only thing I am missing in my Lums are those portions of the color spectrum that my filters do not pass and I really don't want those lines anyway.  I have eliminated shooting lums and now only shoot RGBs unbinned, creating a synlum as per the thread.  I haven't had any regrets so far and, like Rick, shoot with both 16803 and 8300 chips.

For what its worth,

Best,

Jim

[Zocky]

Thnx Jim.